> Focusing on developers, Google also unveiled an impressive way for game developers to apply their own design style to titles on Stadia. It’s a machine learnin

> Focusing on developers, Google also unveiled an impressive way for game developers to apply their own design style to titles on Stadia. It’s a machine learning-based style transfer tool that developers can use to simply drop an image into the video frames of games and have it mimic the style throughout. This is a fascinating testament to the rate at which deep learning can run. If you think back to 2014-2016 (less than 5 years ago), one of the common things people said was that style transfer would be really cool to apply to video games or video, and then the middlebrow dismissal was 'style-transferring a single frame Gatys et al 2014 style takes a Titan hours so [simple math] shows we can't style transfer in realtime for decades, duh'. But, GPUs got better. Deep learning got better. Style transfer got better. Now, here in 2019, realtime style transfer is not just old hat (Facebook launched a realtime style transfer video app for smartphones years ago), but Google is announcing a service to stream countless hours of style-transferred video games to potentially millions of players.